All notable changes to maq
will be documented in this file.
The format is based on Keep a Changelog and this project adheres to Semantic Versioning.
integrated_difference
where the AUC measure is wrong if \bar B exceeds the point at which the curve plateaus. #44
maq
function signature to make budget
an optional argument. The default behavior (budget = NULL
) is to fit the Qini curve up to a maximum spend/unit where each unit that is expected to benefit, is treated. #41
integrated_difference(object.lhs, object.rhs, spend)
for estimating the area between two Qini curves up to some maximum budget spend
. #42
predict.maq
return a standard dense matrix, and remove dependence on the sparse Matrix package. #30
First CRAN beta release (this changelog tracks the R package). The R package currently supports
maq(cate.hat, cost.hat, max.budget, Y.eval, ...)
, where cate.hat
and cost.hat
are CATE and cost estimates obtained via some function learned on a training set, Y.eval
are evaluation scores on a test set (for example inverse-propensity weighted outcomes) and max.budget
is the maximum mean budget/unit to fit the curve on. Setting the option target.with.covariates
to FALSE
yields a baseline Qini curve that can be used to assess the value of treatment targeting based on covariates.average_gain()
(supporting clustered standard errors if fit with clusters).difference_gain()
, yielding standard errors that account for the correlation arising from curves fit on the same evaluation data.predict()
.plot()
.summary(maq.object)
.The Python bindings (source install only) currently supports